Converging the k-mesh


The k-mesh is constructed using the tetrahedron of Bloechl et al., Phys. Rev. B, 1990. It is written to file early on. The number of k-points in the k-mesh is an unphysical parameter that simply needs to be converged. Although it is impossible to give a general guideline, starting with 1000 k-points is a good idea for smaller unit cells. Generally, the number of points needed scales inversely with the volume of the unit cell. Some systems require more points than others. One always needs to check. The more broadened the property of interest (e.g. ELNES as opposed to DOS), the fewer points are necessary. Also, the near edge structure requires more points, whereas more extended structure (e.g., 50-70 eV above threshold) is often converged with just a few k-points.

Generally speaking, the calculations of the potentials requires less accuracy than the calculation of FMS. Just like one usually uses a smaller SCF-radius than the FMS-radius for real-space calculations, it makes sense to use a smaller k-mesh for SCF than for FMS. Therefore, it can be a good strategy to, e.g., set the k-mesh to 200 points, run the potentials calculation, then raise the number of k-points to, e.g., 1000, and then run the FMS calculation. This can save much calculation time. However, there are exceptions; e.g. to calculate the potentials of anatase TiO2 we used at least 500k points, more than needed for fms.

developer's resources